The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents patterns for building workflows that ingest data from untrusted external sources, creating a surface for injection attacks if the data is processed by AI agents or downstream systems.
Ingestion points: webhook_processing.md (receiving HTTP POST bodies) and http_api_integration.md (consuming responses from external REST APIs).
Boundary markers: While the skill suggests validation steps, it does not provide standardized delimiters for isolating untrusted data during interpolation into prompts or commands.
Capability inventory: Described workflows have the capability to execute SQL writes (database_operations.md), perform arbitrary API calls (http_api_integration.md), and trigger AI-driven actions (ai_agent_workflow.md).
Sanitization: The documentation proactively addresses this risk by teaching defensive measures such as SQL parameterized queries to prevent SQL injection and HMAC signature verification using the crypto module for webhooks.
[DYNAMIC_EXECUTION]: The skill provides multiple JavaScript code snippets intended for execution within n8n "Code" nodes, which allow for runtime script evaluation.
Evidence: scheduled_tasks.md provides an example using require('child_process').exec to run system commands (e.g., pg_dump). webhook_processing.md provides snippets using the crypto module for data verification.
Context: This dynamic execution is a standard feature of the n8n platform. The skill provides these as educational templates and encourages safe implementation, such as using environment variables for paths and credentials for authentication.

n8n-workflow-patterns